NVIDIA posted numbers higher than our estimates and our preview, with a clean beat and raise on all metrics, with a significant Vera Rubin ramp ahead that should prove out their contention that NVDA hardware leads in AI factory economics. Best value in semis, remain OW/Top Pick. Target $288

• No big surprises, given elevated expectations, but we continue to like the setup. Market share rhetoric is likely hitting a low point, with Vera Rubin providing an answer to key competitive questions, and understanding that the elevated benchmark weighting presents certain headwinds, but valuation at about half of AI peers really does stand out.

Measuring market share is tricky, and not particularly meaningful. Everyone is supply constrained, and while that brings NVIDIA significant benefits, as the company most aggressively managing the supply chain, it brings challenges as well. $13 bn of sequential revenue growth this quarter is as much revenue as the entire revenue of the 3rd biggest logic company in our coverage, and the company would appear to need most of the capacity for its served nodes next year.

The fundamental debate should be: can NVIDIA deliver on its claim that the lowest cost per token happens on NVIDIA hardware? We think that they have consistently done that, and that they can continue to do that. The goal posts are always moving, and we are constantly hearing about ASICs or competing chips that offer better TCO, but those are invariably comparisons of competitors' new products such as AMD MI455 or TPU v8, with NVIDIA's current products, despite the fact that Vera Rubin will likely have general availability before either of those.

That doesn't mean the comparisons are irrelevant, of course, as $400 bn run rate of AI capital investment cannot be fully satisfied with Vera Rubin for several quarters, so Blackwell vs. other people's next gen products is still an important part of the consideration set. But we believe the NVIDIA truism that lower cost silicon has not been proven to deliver lower cost per tokens, until proven otherwise, and that having the best in class capability is a considerable advantage for customers that want the longest useful life.

The resegmentation tells a compelling story about strong growth above and beyond frontier model developers, though we heard mixed reactions from investors. Our focus will remain on data center growth overall, and we are less concerned about subsegments below that line, but the split between hyperscale and other customers shows that roughly half of the business is outside of the concentrated hyperscale environment and growing nicely. Hyperscale growth of 116% would be even higher without the higher China year ago #s. Concerns that upside came mostly from networking seem mostly irrelevant to us.

Meanwhile, claims of revenue leadership in CY26 server CPUs reverberatesthrough the suddenly overcrowded server CPU trade. The $20 bn target for CPU this year would put them right at the cusp of market leadership - we have Intel data center revenues slightly higher, but that includes some networking, and AMD just below. We do think the $20 bn includes head node CPUs (i.e. delivered on the GPU cards), which counts but is a bit easier of a sale that is not incrementally bad for x86 competitors, but it also has to include meaningful encroachment into standalone server racks for agentic workloads. We note that the company had prior agreements to sell standalone Vera to Coreweave (covered by Keith Weiss) and Meta (covered by Brian Nowak), but then more recently announced standalone shipments to, among others, Oracle (also covered by Keith Weiss). We are excited for this opportunity and think that NVIDIA's strong procurement will put them in a very strong position in a supply constrained environment. We expect to hear more about this at the Computex/GTC keynote in early June.

Other important highlights:

• Significant cash return. Dividends are up to $1 annual run rate - about 0.5% yield - and an $80 bn buyback.
• Vera Rubin tracking for 3Q delivery. Given constant anecdotes of delays, this timing - at the earlier end of prior 2H commentary - should provide some relief
• $119 bn in purchase commitments, up almost $30 bn from last qtr, with $95 bn of that used in production in the next 3 qtrs - which would minimize concerns about supply constraints vs. current #s.
• Gross margins slightly better. While the company has consistently talked about mid 70s gross margins, they have also talked about 75% as a limiter, so guidance of 75% amid component price increases should be a relief.

Details on the Quarter: April revenue of $81.615bn (up 19.8% q/q and 95.1% y/y) was above the Street at $78.907bn and our estimate of $79.264bn. By segment, Data Center revenue of $75.249bn was up 20.8% q/q and 92.4% y/y, within datacenter Hyperscale increased 12% q/q and 115% y/y with ACIE up 31% q/q and 75% y/y, Edge was up 9.6% q/q and 28.7% y/y to $6,369bn. Gross margin of 75.0% was in-line with the Street and 0.1% below our estimate of 75.0% and 75.1%, respectively. Non-GAAP EPS of $1.87 beat both the Street and our estimate of $1.75, and $1.72 respectively.

NVIDIA guided for an $10bn sequential revenue increase;to $91bn at the midpoint for July (up 11.5% q/q and 94.7% y/y), which came in ahead of the Street and our estimate of $87.293bn and $87.880bn, respectively, with data center once again the biggest contributor to growth. The company guided gross margin to 75.0%, compared to the Street at 74.5% and our estimate of 75.1%.

Raising estimates: We are increasing our revenue estimates for FY27 and FY28, now modeling 82.0% y/y growth and 52.4% y/y growth, respectively, vs 76.3% and 54.4% prior. For the Oct quarter, we move revenue from $99.015bn to $102.346bn and non-GAAP EPS from $2.25 to $2.34. For FY27, our revenue and non-GAAP EPS move from $380.591bn/$8.61 to $393.005.450/$8.98, and FY28 comes up to $598.809bn/$13.15 from $587.450bn/$13.11 previously.

Thoughts on the stock: Near term expectations have been high, given an obviously strong spending pattern, but market share rhetoric has been negative, given ASIC forecasts and performance claims. The reality is that the world is short compute, and that anyone who can supply compute with memory attached to it is generating material interest. But in 2026, and again in 2027, some of NVIDIA's strongest growth customers will be hyperscalers who also use ASICs, which should be pretty telling. Vera Rubin is a strong answer to that. With the sudden surge of interest in CPUs, the company's claim of revenue share leadership is quite interesting, and we'll look to get a better sense of the company's penetration into agentic standalone CPUs in the coming weeks.

We understand that the company's index benchmark can be a natural hurdle, and that perceptions of share loss may take time to diminish. But the relative valuation gap is just too wide to ignore, and to us the company's position as the most important AI chip supplier remains unassailable.

In terms of our PT, we are leaving our methodology unchanged from our prior of ~22x MW EPS ests for CY27 (FY28). On our new $13.08 MW EPS that brings the PT up to $288 vs $285 previously. ~22 MW EPS is in-line with the broader market and a discount to compute semis peers (AMD/AVGO/INTC) as high marketshare and gross margins leave limited levers for multiple expansion in the near term.